Large-scale Approximate Kernel Canonical Correlation Analysis


MATLAB package for Kernel Nonlinear Orthogonal Iterations (KNOI)

(C) 2015 by Weiran Wang and Karen Livescu.

Download the package here.

This Matlab code implements the KNOI algorithm described in the paper:
Weiran Wang and Karen Livescu.
Large-scale Approximate Kernel Canonical Correlation Analysis.

Quick start:
- demo.m: demonstrates the application of linear/approximate kernel CCAs on 
  left/right halves of MNIST images as the two views. Use 60K/10K samples 
  for training/validation.

List of functions:
- KNOI.m: the Kernel Nonlinear Orthogonal Iterations algorithm.
- KNOI_forward: applies KNOI projection mapping to test samples.
- linCCA.m: the linear CCA algorithm.
- randKCCA.m: the randomized kernel CCA algorithm, proposed in 
  D. Lopez-Paz, S. Sra, A. Smola, Z. Ghahramani, and B. Schoelkopf.
  Randomized Nonlinear Component Analysis.
  The 31st International Conference on Machine Learning (ICML) 2014.
- randKCCA_forward.m: applies randKCCA projection mapping to test samples.
- createMNIST.m: generates the halved MNIST images used in demo.m (the 
  random seed used to generate our data is saved in demoseed.mat, so that 
  you can achieve exactly the same result).

External packages/data:
- mnist_all.mat: all MNIST images in MATLAB format can be downloaded from 
  Sam Roweis's webpage http://www.cs.nyu.edu/~roweis/data.html.

Send email to Weiran Wang.

TTI-Chicago | Weiran Wang's Home Page